Update to PromptingCraft.com

Changed files

Choose a file to see its text changes. Binary or large files show metadata only.

.claude-plugin/plugin.json →

README.md →

plugin.json →

skills/promptingcraft/SKILL.md →

skills/promptingcraft/references/knowledge.json →

skills/promptingcraft/references/sources.json →

plugin.json

− removed · + added

--- before
+++ after
@@ -1,8 +1,8 @@
 {
   "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
   "name": "promptingcraft",
-  "version": "0.3.0",
-  "description": "Turn development discussions into execution-ready prompts for coding agents such as Codex, Claude Code and Cursor, with source-backed official guidance. Developed by the dondego.com team.",
+  "version": "0.4.0",
+  "description": "Write, improve and review execution-ready prompts and task handoffs for coding agents such as Codex, Claude Code and Cursor, using source-backed guidance. Developed by the dondego.com team.",
   "author": {
     "name": "Aleksandr Prokofyev",
     "url": "https://promptingcraft.com"
@@ -28,7 +28,7 @@
       "interface": {
         "displayName": "PromptingCraft.com",
         "shortDescription": "Source-backed agent prompts",
-        "longDescription": "PromptingCraft.com helps AI assistants turn a development discussion into one clear, execution-ready prompt for a coding agent such as Codex, Claude Code or Cursor.\n\nIt identifies the task contract (requirements, decisions already made, permissions and open questions), checks applicable official guidance from AI developers, writes a draft, critiques it against the contract, and returns one final copyable prompt. Optional ideas stay optional, unresolved choices stay open, and no new permissions are added.\n\nIt can also review an existing prompt: it points out unnecessary instructions and changes in scope, and when a prompt is already suitable it says so and leaves it unchanged.\n\nModel-specific recommendations come only from official OpenAI and Anthropic documentation, and Cursor behavior from Cursor's documentation. A public source registry records where each recommendation comes from, and the project's own methodology is labeled separately from official guidance.\n\nThe skill works on its own. The optional PromptingCraft.com guidance server (mcp.promptingcraft.com) adds read-only access to the same ledger: it lists official guidance for a target agent, model family and topic, returns a record with its source, limits and verification date, and checks whether an official source page changed since it was verified. Its tools take only fixed categories and identifiers as input and do not ask for task text.\n\nChecking current guidance depends on the tools and access available in the conversation. When the server or live retrieval is unavailable, PromptingCraft.com says so and states whether it used other official sources, dated cached guidance or none.\n\nPromptingCraft.com writes and reviews prompts only. It does not run the development task, edit code or grant permissions, and it does not guarantee better downstream results. It needs no account, credentials or setup, and no PromptingCraft.com service receives your prompts. Evaluation methods, results and limits are published at promptingcraft.com/benchmarks.\n\nPromptingCraft.com is developed independently and is not affiliated with or endorsed by OpenAI, Anthropic or Cursor.\n\nDeveloped by the dondego.com team.",
+        "longDescription": "PromptingCraft.com helps you write, review and improve prompts for coding agents such as Codex, Claude Code and Cursor. Turn a development discussion into clear, execution-ready instructions for an AI developer.\n\nIt identifies the task contract (requirements, decisions already made, permissions and open questions), checks applicable official guidance from AI developers, writes a draft, critiques it against the contract, and returns one final copyable prompt. Optional ideas stay optional, unresolved choices stay open, and no new permissions are added.\n\nIt can also review an existing prompt: it points out unnecessary instructions and changes in scope, and when a prompt is already suitable it says so and leaves it unchanged.\n\nBundled model-specific records include official OpenAI and Anthropic guidance and scoped DeepSeek and Kimi API tool-use guidance; Cursor behavior comes from Cursor documentation. A public source registry records where each recommendation comes from, and the project's own methodology is labeled separately from official guidance.\n\nThe skill works on its own. The optional PromptingCraft.com guidance server (mcp.promptingcraft.com) adds read-only access to its deployed guidance ledger: it lists official guidance for a target agent, model family and topic, returns a record with its source, limits and verification date, and checks whether an official source page changed since it was verified. Its tools take only fixed categories and identifiers as input and do not ask for task text.\n\nChecking current guidance depends on the tools and access available in the conversation. When the server or live retrieval is unavailable, PromptingCraft.com says so and states whether it used other official sources, dated cached guidance or none.\n\nPromptingCraft.com writes and reviews prompts only. It does not run the development task, edit code or grant permissions, and it does not guarantee better downstream results. It needs no PromptingCraft.com account or credentials; installation and connection availability depend on the host. The guidance tools take only fixed categories and identifiers, and no PromptingCraft.com service receives your prompts. Evaluation methods, results and limits are published at promptingcraft.com/benchmarks.\n\nPromptingCraft.com is developed independently and is not affiliated with or endorsed by OpenAI, Anthropic or Cursor.\n\nDeveloped by the dondego.com team.",
         "developerName": "Aleksandr Prokofyev",
         "category": "Developer Tools",
         "capabilities": [
@@ -45,9 +45,9 @@
         "privacyPolicyURL": "https://promptingcraft.com/privacy",
         "termsOfServiceURL": "https://promptingcraft.com/terms",
         "defaultPrompt": [
-          "Turn our discussion into an execution-ready Codex prompt.",
-          "Review this Claude Code task without changing its scope.",
-          "Apply relevant official guidance and preserve my requirements."
+          "Write a Codex prompt to fix this bug without changing the API.",
+          "Help me tell Claude Code what to implement from our discussion.",
+          "Review my Cursor agent prompt for missing requirements."
         ],
         "brandColor": "#15191F",
         "brandColorDark": "#E3B052",
@@ -61,7 +61,7 @@
           "positive": [
             {
               "description": "Find official guidance for a Codex prompt",
-              "prompt": "Using PromptingCraft.com, which official OpenAI guidance on tool use applies to a prompt I am writing for Codex?",
+              "prompt": "Which official OpenAI tool-use guidance applies to a Codex prompt I am writing?",
               "tools_triggered": "search_guidance",
               "expected_behavior": "Lists curated OpenAI records on tool use (for example oai-tools and oai-structured) with their source URLs, labels and verification dates. It does not present PromptingCraft.com methodology as official guidance."
             },
@@ -110,7 +110,7 @@
         "commerce_description": "PromptingCraft.com is free. It sells nothing and processes no payments."
       },
       "publication": {
-        "release_notes": "Adds the PromptingCraft.com guidance server (https://mcp.promptingcraft.com/mcp): three read-only tools that search the curated official-guidance ledger, return a record with its provenance, and check whether an official source changed since it was verified. The skill uses the server when it is available; otherwise it says so and falls back to other official sources or its bundled dated references. No account is needed. Developed by the dondego.com team."
+        "release_notes": "Adds bundled, scoped DeepSeek and Kimi API tool-use guidance with source provenance. The deployed server at https://mcp.promptingcraft.com/mcp may expose an earlier ledger and schema; the skill handles that difference. It uses the server when it is available; otherwise it says so and falls back to other official sources or its bundled dated references. No account is needed. Developed by the dondego.com team."
       },
       "apps": "./.app.json"
     }